14  BA Sim Study

This page summarizes the BA kernel simulation study results saved as per-replicate .rds files under:

Each replicate contains truth, estimates, counts, and runtimes (see str(ba_rep_003)).

14.1 Quick summary

tab <- summary_table(df)
tab
# A tibble: 6 × 7
  T_end     n keep_rate n_events_mean n_events_sd sim_s_mean fit_s_mean
  <dbl> <int>     <dbl>         <dbl>       <dbl>      <dbl>      <dbl>
1     5   350     0.954          81.5        14.6      0.297       1.46
2    10   350     0.963         180.         24.3      0.530       2.30
3    25   350     0.994         476.         41.2      1.88        6.46
4    50   350     0.997         975.         62.8      5.86       16.2 
5    75   350     0.997        1476.         79.5     12.4        35.3 
6   100   350     0.997        1973.         94.2     19.4        50.7

14.2 Cap-Exceeded Replicates

There were a few replicates which “exploded”, and ended up with unrealistic estimates of K & beta. I decided to just filter these out, as it was making the plots unreadable.

Chose to filter out any \(K > 20\), \(\beta > 20\) (arbitrarily).

df %>%
  summarise(
    n_total = n(),
    n_keep  = sum(keep, na.rm = TRUE),
    n_dropped_caps = sum(drop_reason == "cap_exceeded", na.rm = TRUE)
  )
# A tibble: 1 × 3
  n_total n_keep n_dropped_caps
    <int>  <int>          <int>
1    2100   2066             34
df_non_filtered <- df
df <- df %>% filter(keep)

14.3 Parameter estimates vs T_end

Dashed line = true value.

plot_estimates_vs_T(df, keep_only = TRUE)

Log scale:

plot_estimates_vs_T(df, keep_only = TRUE, log_scale = TRUE)

14.4 Counts vs T_end

plot_counts_vs_T(df)

14.5 Keep rate vs T_end

plot_keep_rate_vs_T(df_non_filtered)

14.6 RMSE decay vs T_end

plot_rmse_vs_T(df)

14.7 Runtime vs T_end

plot_runtime_vs_T(df)